Study on Control of Facial Expressions of Face Robot and Their Informatic Characterisitics

(1)人脸机器人由头架结构、18个面部肌肉FMA驱动器、硅橡胶脸部皮肤、眼、口、鼻和颈部结构组成，具有18个自由度来模拟人脸运动。该人脸机器人的大小比正常人脸大20%。根据心理学中常用的动作单元的种类和强度，确定了人脸六种基本表情的运动方式，并建立了人脸表情的控制算法。(2)对人脸机器人上显示的六种基本人脸表情进行了视觉测试，获得了90%左右的正确率(除恐惧表情外)。(3)研制了用于实时表情运动的小型气缸活塞式执行器，并将其中的18个执行器应用到人脸机器人中，通过在每个执行器中设置反馈控制，建立了控制算法，实现了实时的表情控制人脸机器人上的面部表情运动。(4)通过使用Transputer系统和2个CCD摄像头，成功地对眼睛、眉毛和嘴巴的运动进行了实时测量。一次动态人脸图像测量所需的时间约为100ms，对于我们的面部表情动态识别来说已经足够短了。(5)将Transputer系统获得的动态人脸图像数据应用到由典型面部表情训练的神经网络中，并对人脸机器人的面部表情进行了识别测试，获得了相当高的准确率(约80%的正确率)。(6)未来的工作将是利用面部表情来研究人与人脸机器人之间心理信息交换的交互特性。

(1) The face robot consitutes of a head-frame structure, 18 facial muscle FMA actuators, silicone-rubber face skin, eyes, mouth, nose and neck structure and has 18 degree-of-freedom to mimic the movement of human face motion. The size of the face robot is 20% larger tahn that of normal human face. The motion of face components for expressiong 6 basic facial expressions are determined according to the kind and strength of Action Units used in psychology and control algorithm for the facial expressions has been established.(2) The 6 basic facial expressions displayd on the face robot were evaluated by visual test and obtained about 90% correct expression except for "Fear" expression.(3) Small size air-cylinder-piston type actuator was developed for real-time motion of facial expression and 18 of this actuators were implemented into the face robot and control algorithm was established by using a feedback control in each actuator for realizing a real-time motion of facial expressions on the face robot.(4) By using the transputer system and 2 CCD cameras, the real-time measurement of motions in eyes, eye brows and mouth was successfully performed. The time needed for one dynamic facial image measurement was about 100 ms which is sufficently short for our dynamic recognition of facial expressions.(5) The dynamic facial image data obtained by the transputer system were applied to the neural network trained by typical facial expressions and recognition tests were undertaken to evaluate the performance of facial expressions of the face robot and obtained a rather high rate (about 80% correct ratio).(6) Some of future works will be to investigate the interactive characteristics of psychological information exchange between human and face robot by using facial expressions.